Help me with Phase Only Image Correlation!

Hello, I am attempting to correlate an image drawn from a video with a cropped out portion of the same image. We are having dimension problems that wont seem to go away. I believe that there may be some underlying problem that I am not recognizing. Any help is appreciated, I have posted the code below. Unfortunately the Mp4 file we used is too large to attach, but this code should operate the same for any video file.

function[] = Again(~)

prompt = 'Name of file?\n'; %Prompt user to input a video file.

result = input(prompt);

%Step through the video frame by frame. We chose 58 as an arbitrary point %to stop where we know a good variety of objects are present.

videoFReader = vision.VideoFileReader(result);

videoPlayer = vision.VideoPlayer;

j = 0;

while j<58;

%while ~isDone(videoFReader)

    videoFrame = step(videoFReader);
    step(videoPlayer,videoFrame);
    j=j+1;
    i=step(videoFReader); %Save the image as variable i
    hold on
end
%imtool(i);

%Crop out a section of image(i) and assign it variable h. Here we did it in %two seperate ways, just for fun...

%h=imcrop(i,[1016 488 219 157]);

h = i(488:645,1016:1235,3);

h(720,1280,3)= 0; %Ensure image dimensions of i and h match

imshow (h); %Now the hat is red?

figure, imshow (i);

%Here we used the ndims, length, & numel functions to determine if each image %has the desired matrix dimensions and more importatntly that they are %identicle.

%ndims(i)

%length(i)

%numel(i)

%ndims(h)

%length(h)

%numel(h)

%In our first attempt to correlate two images we chose to first attempt to %use both magnitude and phase. However now we are attempting to us only the %phase to correlate the images and create a mesh to show any matches. We %have included both attempts although neither work. Each returns similar %errors about the matrix dimensions of the two files.

%********************* Attemp to coorrelate #1*****************************

%To ensure the the cropped image has the same dimensions %as image i, we include the 720,1280

I = fft2(i);

H = fft2(h,720,1280);

%Here is what we beleive to be two functional ways to extract the phase %information the fourier transfor of an image. We chose the latter for %simplicity.

%F_Phase = cos(angle(I)) + j*(sin(angle(I)));

%F_Phase2 = cos(angle(H)) + j*(sin(angle(H)));

F_Phase = exp(1j*angle(I));

F_Phase2 = exp(1j*angle(H));

%This is what we beleive would be the way to correlate two images based on %phase and magnitude information.

%conjH=conj(H);

%R = I .* conjH;

%We have read that correlation is like convlution but rotated 180 degrees, % and mathmatically it makes sense. However this is where the error occur. % The error we get regularily is that "The Matrix sizes must be the same" % despite the images being the same dimensions.

R = F_Phase .* (rot90(F_Phase2,2));

%Here we take the inverse fourier transform and set it to real values so %that we may view the results using the mesh function.

res=real(ifft2®);

%size(res)

mesh(res);

%*************************************************************************

%Release the video player and reader.

release(videoPlayer);

release(videoFReader);

end

Answer 1

[Dima Lisin]

Image Processing Toolbox now includes the imregcorr function, which does phase correlation.